Xirp by Spotify

Import a project. Run a session. Switch agents mid-task with full context carried over. Xirp lets you start a session in Claude Code then swap to OpenAI's Codex without losing a thing. The conversation history, the findings, the context. It all transfers. Different agent, same session. Try now https://hubs.ly/Q04y2GYV0

💫 Wrapping up the week with a look back at Xirp releases (v0.32–v0.35) — a new coding agent, CLI improvements, and a round of workflow polish. - Pi coding agent — Xirp now supports Pi! This is alongside Claude Code, OpenAI's Codex, Gemini, and Cursor. Install it from the agent picker or Settings. - Worktree search — filter a project's worktrees by branch name, folder, or active session when the list gets long - Linked shell terminals from the CLI — Xirp session new-terminal creates a plain shell linked to a running session, with auto-detection of the parent - Repository insight during onboarding — Xirp scans the repo when you set up your first task and suggests starter goals based on languages, health, and activity - Restore from backup — if the local database can't start, Xirp offers to recover from a saved backup instead of starting fresh - Pinnable toolbar actions — choose which session actions stay visible; the rest move to a more menu Also worth noting: sessions with background commands no longer end prematurely, CJK input no longer double-submits on Enter, and a new notification preferences panel lets you filter or batch what reaches you. Full changelog → https://hubs.li/Q04xTcgT0

How much did your AI agents cost you yesterday? What about the day before? With Xirp, you can now see costs by agent, by model, and by day, broken down across your coding agents and every model you're running. Hover over any day and see exactly where your tokens went. The first step to managing AI tooling spend is knowing what you're spending.  Xirp gives you that knowledge. Try now https://hubs.li/Q04xKymQ0

One of the most common requests from Xirp users: more choice in coding agents and models.  Today we're taking a step in that direction. Xirp by Spotify now supports Pi, an open-source coding agent with access to a broad range of models. Same workflow, dozens of concurrent sessions, way more choice.  Try it out: https://hubs.li/Q04xx1LS0

Here's a glimpse at this week's Xirp releases (v0.26 through v0.31), headlined by a command-line tool and browser panel annotations. → CLI tool — `xirp` now installs alongside the desktop app. Import sessions from your preferred agents, list running sessions, send messages, and upload transcripts, all from a terminal or from inside a coding agent. → Browser panel annotations — click elements on a page to highlight them, leave comments, and send all annotations to the coding agent in one prompt. Great for UI review workflows. → Cost alerts — set a daily spend threshold under Usage & Costs. Xirp warns you when a session approaches or crosses it, so runaway agents don't surprise you at the end of the week. → Mermaid diagrams — Markdown files with Mermaid code blocks now render the diagram in the file viewer instead of showing raw text. → Portal connection banner — sessions started before Portal was linked now offer a one-click restart to pick up Portal tools. → Interrupted session recovery — sessions lost to infrastructure failures (crashed terminal server, machine reboot) are now marked as interrupted instead of silently shown as completed, and can be resumed. Check out the full changelog ⤵️ https://hubs.li/Q04xgMHD0 Try Xirp free ⤵️ https://hubs.li/Q04xgMJT0

Your AI agents are running up a tab you're not tracking. Every session starts from zero, scanning, guessing, orienting, before a single useful line ships. Multiply that across your team and it becomes one of your biggest AI tooling line items. We measured the hidden tax at Spotify. Then we fixed it. Here's how https://hubs.li/Q04x2RtF0

One prompt. Zero exploring. 🎯 We started a OpenAI Codex session from a Portal by Spotify workspace. The agent's first move? Pull service ownership, dependencies, and architecture automatically. No "what repo is this?" No five minutes of codebase spelunking. The context came to the agent, not the other way around. Portal workspace → Xirp session → grounded from the first response. Your context. Your codebase. Your agent already knows what it's looking at. Try it → https://hubs.li/Q04wYHns0

7 releases landed in Xirp this past week (v0.20 through v0.25). Here's the headline: You can now hover over an agent's response, select specific passages, annotate each one, and send them back as a single follow-up prompt. It turns "no, not that part" into something precise. More in: Bulk worktree cleanup — a new Cleanup button finds stale worktrees across a project, shows age, merge status, and uncommitted work, and lets you remove them in one pass Blocked workflow status — sessions can now be marked blocked alongside backlog, in progress, in review, and done Visual refresh — redesigned header, menus, and panel layouts, with independent app and terminal themes so dark terminal + light UI is finally a thing Security — Portal OAuth tokens moved from plaintext on disk to OS secure storage Data safety — crash recovery now preserves your local database instead of silently starting fresh Check out the full changelog at https://hubs.li/Q04wLRp-0 or visit https://hubs.li/Q04wLRp30 to try Xirp for yourself.

One thing we learned building developer platforms at @Spotify: The real test starts when engineers begin using and relying on them. We released @Xirp publicly last month, and since then, the feedback has started flowing. The team has shipped eight beta releases, and you’ve been telling us what works well and where the gaps remain. A lot of the work has focused on reliability: better sign-in and session recovery, more accurate session states and notifications, clearer startup failures, and fixes for the edge cases that show up once engineers start relying on a tool in even more diverse workflows. For me, that reliability work is crucial because every engineer should be able to trust Xirp every day as we do at Spotify. Backstage grew inside Spotify because our developers were also its customers. They used it every day, told us where it got in their way, and helped make it better. Xirp is going through that same process now, only at an even faster pace. Engineers and technical leads outside Spotify are now using Xirp in their own workflows, giving us a much broader range of experience to learn from. We value all of the feedback you’re sharing with us. Keep it coming. See everything that’s changed: https://lnkd.in/dbhKeSAS

A few weeks in, we’ve already shipped eight beta releases for @Xirp, so here’s a quick update on what’s changed. One of the problems we talked about at launch was coordination: engineers running multiple coding agents across different projects, branches, and repos, each with its own setup and context. Since then, we’ve been hearing from engineers and technical leads using Xirp in their own workflows, and we’ve been closing some of those gaps they’ve flagged. We’ve recently added Cursor alongside Claude Code, Codex, and Gemini. Sessions can be grouped by worktree, so it’s easier to see which project and branch each one belongs to. And when a session spans multiple projects, each agent now looks for the correct instruction files for its own project. Agent setup and context are more consistent now, too. Codex gets the same launch context as Claude sessions, and Codex sessions launched from Portal now receive the workspace tools and project context they were previously missing. Claude Code and Codex now use their vendors’ official installers, and the new Doctor page checks your coding agents, tmux, and agent hooks, then shows you what failed and how to fix it. We’ve also fixed issues around sessions starting in the wrong directory, Codex sending false waiting notifications, startup failures, and other reliability gaps that get in the way when you’re running multiple agents in parallel. All of these changes are focused on the same problem: When you’re running more agents across more projects and tools, Xirp should make it easier to know where each session is running, what context it has, and when it actually needs your attention. That’s what we’re continuing to work on with every change we ship for Xirp. Keep the feedback coming. See all the changes here: https://lnkd.in/dcCQysAh